Machine Learning Predicts Cerebral Vasospasm in Subarachnoid Hemorrhage Patients.

Background: Cerebral vasospasm (CV) is a feared complication occurring in 20-40% of patients following subarachnoid hemorrhage (SAH) and is known to contribute to delayed cerebral ischemia. It is standard practice to admit SAH patients to intensive care for an extended period of vigilant, resource-intensive, clinical monitoring. We used machine learning to predict CV requiring verapamil (CVRV) in the largest and only multi-center study to date. Methods: SAH patients admitted to UCLA from 2013-2022 and a validation cohort from VUMC from 2018-2023 were included. For each patient, 172 unique intensive care unit (ICU) variables were extracted through the primary endpoint, namely first verapamil administration or ICU downgrade. At each institution, a light gradient boosting machine (LightGBM) was trained using five- fold cross validation to predict the primary endpoint at various timepoints during hospital admission. Receiver-operator curves (ROC) and precision-recall (PR) curves were generated. Results: A total of 1,750 patients were included from UCLA, 125 receiving verapamil. LightGBM achieved an area under the ROC (AUC) of 0.88 an average of over one week in advance, and successfully ruled out 8% of non-verapamil patients with zero false negatives. Minimum leukocyte count, maximum platelet count, and maximum intracranial pressure were the variables with highest predictive accuracy. Our models predicted "no CVRV" vs "CVRV within three days" vs "CVRV after three days" with AUCs=0.88, 0.83, and 0.88, respectively. For external validation at VUMC, 1,654 patients were included, 75 receiving verapamil. Predictive models at VUMC performed very similarly to those at UCLA, averaging 0.01 AUC points lower. Conclusions: We present an accurate (AUC=0.88) and early (>1 week prior) predictor of CVRV using machine learning over two large cohorts of subarachnoid hemorrhage patients at separate institutions. This represents a significant step towards optimized clinical management and improved resource allocation in the intensive care setting of subarachnoid hemorrhage patients.

Lumbar Spinal Canal Segmentation in Cases with Lumbar Stenosis Using Deep-U-Net Ensembles.

BACKGROUND: Narrowing of the lumbar spinal canal, or lumbar stenosis (LS), may cause debilitating radicular pain or muscle weakness. It is the most frequent indication for spinal surgery in the elderly population. Modern diagnosis relies on magnetic resonance imaging and its inherently subjective interpretation. Diagnostic rigor, accuracy, and speed may be improved by automation. In this work, we aimed to determine whether a deep-U-Net ensemble trained to segment spinal canals on a heterogeneous mix of clinical data is comparable to radiologists' segmentation of these canals in patients with LS. METHODS: The deep U-nets were trained on spinal canals segmented by physicians on 100 axial T2 lumbar magnetic resonance imaging selected randomly from our institutional database. Test data included a total of 279 elderly patients with LS that were separate from the training set. RESULTS: Machine-generated segmentations (MA) were qualitatively similar to expert-generated segmentations (ME1, ME2). Machine- and expert-generated segmentations were quantitatively similar, as evidenced by Dice scores (MA vs. ME1: 0.88 ± 0.04, MA vs. ME2: 0.89 ± 0.04), the Hausdorff distance (MA vs. ME1: 11.7 mm ± 13.8, MA vs. ME2: 13.1 mm ± 16.3), and average surface distance (MAvs. ME1: 0.18 mm ± 0.13, MA vs. ME2 0.18 mm ± 0.16) metrics. These metrics are comparable to inter-rater variation (ME1 vs. ME2 Dice scores: 0.94 ± 0.02, the Hausdorff distances: 9.3 mm ± 15.6, average surface distances: 0.08 mm ± 0.09). CONCLUSION: We conclude that machine learning algorithms can segment lumbar spinal canals in LS patients, and automatic delineations are both qualitatively and quantitatively comparable to expert-generated segmentations.

Upper lumbar spine far lateral disc herniations masquerading as peripheral nerve sheath tumors: illustrative cases.

BACKGROUND: Migratory disc herniations can mimic neoplasms clinically and on imaging. Far lateral lumbar disc herniations usually compress the exiting nerve root and can be challenging to distinguish from a nerve sheath tumor due to the proximity of the nerve and characteristics on magnetic resonance imaging (MRI). These lesions can occasionally present in the upper lumbar spine region at the L1-2 and L2-3 levels. OBSERVATIONS: The authors describe 2 extraforaminal lesions in the far lateral space at the L1-2 and L2-3 levels, respectively. On MRI, both lesions tracked along the corresponding exiting nerve roots with avid postcontrast rim enhancement and edema in the adjacent muscle tissue. Thus, they were initially concerning for peripheral nerve sheath tumors. One patient underwent fluorodeoxyglucose positron emission tomography-computed tomography (FDG PET-CT) screening and demonstrated moderate FDG uptake on PET-CT scan. In both cases, intraoperative and postoperative pathology revealed fibrocartilage disc fragments. LESSONS: Differential diagnosis for lumbar far lateral lesions that are peripherally enhancing on MRI should include migratory disc herniation, regardless of the level of the disc herniations. Accurate preoperative diagnosis can aid in decision making for management, surgical approach, and resection.

Structural Changes in the Cervicomedullary Junction in Adult Chiari Patients.

OBJECTIVE: To assess volumetric changes in the spinal cord at the cervicomedullary junction, diameter of the cervicomedullary cord, and width of the brainstem following posterior fossa decompression (PFD). METHODS: A retrospective analysis of adult patients with Chiari malformation who underwent PFD was performed. Segmentations were done on clinical quality T2-weighted cervical magnetic resonance images obtained before and after decompression using ITK-SNAP. Volumes of neural tissue within the cervicomedullary junction were evaluated from 10 mm cranial to the medullary beak to the cervical spinal cord at the level of the caudal endplate of the second cervical vertebra. The diameter of the cervicomedullary cord was calculated perpendicular to the spinal cord. The width of the brainstem was measured perpendicular to the clivus at the level of the basion. RESULTS: Twenty adult patients, a mean age of 49.55 years, were included. The cervical cord increased in volume by 13 mm3 to 338 mm3, with an average increase of 155 mm3 (P-value of 0.00002). The diameter of the cervicomedullary cord increased 10.30% 7 mm superior to the beak (P-value of 0.00074), 11.49% at the apex of the beak (P-value of 0.00082), 8.29% 7 mm inferior to the beak (P-value of 0.00075), and the brainstem increased 14.46% perpendicular to the clivus (P-value of 0.00109). The spinal cord at the inferior aspect of the C3 vertebra changed insignificantly (P-value of 0.10580). CONCLUSION: The volume of the cervical cord at the cervical-medullary junction, width of the cervicomedullary cord, and diameter of the brainstem increase following PFD.

Imaging Biomarker Development for Lower Back Pain Using Machine Learning: How Image Analysis Can Help Back Pain.

State-of-the-art diagnosis of radiculopathy relies on "highly subjective" radiologist interpretation of magnetic resonance imaging of the lower back. Currently, the treatment of lumbar radiculopathy and associated lower back pain lacks coherence due to an absence of reliable, objective diagnostic biomarkers. Using emerging machine learning techniques, the subjectivity of interpretation may be replaced by the objectivity of automated analysis. However, training computer vision methods requires a curated database of imaging data containing anatomical delineations vetted by a team of human experts. In this chapter, we outline our efforts to develop such a database of curated imaging data alongside the required delineations. We detail the processes involved in data acquisition and subsequent annotation. Then we explain how the resulting database can be utilized to develop a machine learning-based objective imaging biomarker. Finally, we present an explanation of how we validate our machine learning-based anatomy delineation algorithms. Ultimately, we hope to allow validated machine learning models to be used to generate objective biomarkers from imaging data-for clinical use to diagnose lumbar radiculopathy and guide associated treatment plans.

High-Dose Rate Interstitial Spine Brachytherapy Using an Intraoperative Mobile Computed Tomography-Guided Surgical Navigation System.

BACKGROUND: Up to 15% of previously irradiated metastatic spine tumors will progress. Re-irradiation of these tumors poses a significant risk of exceeding the radiation tolerance to the spinal cord. High-dose rate (HDR) brachytherapy is a treatment alternative. OBJECTIVE: To develop a novel HDR spine brachytherapy technique using an intraoperative computed tomography-guided navigation (iCT navigation). METHODS: Patients with progressive metastatic spine tumors were included in the study. HDR brachytherapy catheters were placed under iCT navigation. CT-based planning with magnetic resonance imaging fusion was performed to ensure conformal dose delivery to the target while sparing normal tissue, including the spinal cord. Patients received single fraction radiation treatment. RESULTS: Five patients with thoracolumbar tumors were treated with HDR brachytherapy. Four patients previously received radiotherapy to the same spinal level. Preimplant plans demonstrated median clinical target volume (CTV) D90 of 116.5% (110.8%-147.7%), V100 of 95.7% (95.5%-99.6%), and Dmax of 8.08 Gy (7.65-9.8 Gy) to the spinal cord/cauda equina. Postimplant plans provided median CTV D90 of 113.8% (93.6%-120.1%), V100 of 95.9% (87%-99%), and Dmax of 9.48 Gy (6.5-10.3 Gy) to cord/cauda equina. Patients who presented with back pain (n = 3) noted symptomatic improvement at a median follow-up of 22 d after treatment. Four patients demonstrated local tumor control of spinal metastatic tumor at a median follow-up of 92 d after treatment. One patient demonstrated radiographic evidence of local tumor progression 2.7 mo after treatment. CONCLUSION: HDR spine brachytherapy with iCT navigation is a promising treatment alternative to induce local tumor control and reduce pain symptoms associated with metastatic spine disease.

Perioperative dual antiplatelet therapy for patients undergoing spine surgery soon after drug eluting stent placement.

BACKGROUND: Performing emergent spinal surgery within 6 months of percutaneous placement of drug-eluting coronary stent (DES) is complex. The risks of spinal bleeding in a "closed space" must be compared with the risks of stent thrombosis or major cardiac event from dual antiplatelet therapy (DAPT) interruption. METHODS: Eighty relevant English language papers published in PubMed were reviewed in detail. RESULTS: Variables considered regarding surgery in patients on DAPT for DES included: (1) surgical indications, (2) percutaneous cardiac intervention (PCI) type (balloon angioplasty vs. stenting), (3) stent type (drug-eluting vs. balloon mechanical stent), and (4) PCI to noncardiac surgery interval. The highest complication rate was observed within 6 weeks of stent placement, this corresponds to the endothelialization phase. Few studies document how to manage patients with critical spinal disease warranting operative intervention within 6 months of their PCI for DES placement. CONCLUSION: The treatment of patients requiring urgent or emergent spinal surgery within 6 months of undergoing a PCI for DES placement is challenging. As early interruption of DAPT may have catastrophic consequences, we hereby proposed a novel protocol involving stopping clopidogrel 5 days before and aspirin 3 days before spinal surgery, and bridging the interval with a reversible P2Y12 inhibitor until surgery. Moreover, postoperatively, aspirin could be started on postoperative day 1 and clopidogrel on day 2. Nevertheless, this treatment strategy may not be appropriate for all patients, and multidisciplinary approval of perioperative antiplatelet therapy management protocols is essential.

Updated external validation of the SORG machine learning algorithms for prediction of ninety-day and one-year mortality after surgery for spinal metastasis.

BACKGROUND CONTEXT: Surgical decompression and stabilization in the setting of spinal metastasis is performed to relieve pain and preserve functional status. These potential benefits must be weighed against the risks of perioperative morbidity and mortality. Accurate prediction of a patient's postoperative survival is a crucial component of patient counseling. PURPOSE: To externally validate the SORG machine learning algorithms for prediction of 90-day and 1-year mortality after surgery for spinal metastasis. STUDY DESIGN/SETTING: Retrospective, cohort study PATIENT SAMPLE: Patients 18 years or older at a tertiary care medical center treated surgically for spinal metastasis OUTCOME MEASURES: Mortality within 90 days of surgery, mortality within 1 year of surgery METHODS: This is a retrospective cohort study of 298 adult patients at a tertiary care medical center treated surgically for spinal metastasis between 2004 and 2020. Baseline characteristics of the validation cohort were compared to the derivation cohort for the SORG algorithms. The following metrics were used to assess the performance of the algorithms: discrimination, calibration, overall model performance, and decision curve analysis. RESULTS: Sixty-one patients died within 90 days of surgery and 133 died within 1 year of surgery. The validation cohort differed significantly from the derivation cohort. The SORG algorithms for 90-day mortality and 1-year mortality performed excellently with respect to discrimination; the algorithm for 1-year mortality was well-calibrated. At both postoperative time points, the SORG algorithms showed greater net benefit than the default strategies of changing management for no patients or for all patients. CONCLUSIONS: With an independent, contemporary, and geographically distinct population, we report successful external validation of SORG algorithms for preoperative risk prediction of 90-day and 1-year mortality after surgery for spinal metastasis. By providing accurate prediction of intermediate and long-term mortality risk, these externally validated algorithms may inform shared decision-making with patients in determining management of spinal metastatic disease.

Predicting Spinal Surgery Candidacy From Imaging Data Using Machine Learning.

BACKGROUND: The referral process for consultation with a spine surgeon remains inefficient, given a substantial proportion of referrals to spine surgeons are nonoperative. OBJECTIVE: To develop a machine-learning-based algorithm which accurately identifies patients as candidates for consultation with a spine surgeon, using only magnetic resonance imaging (MRI). METHODS: We trained a deep U-Net machine learning model to delineate spinal canals on axial slices of 100 normal lumbar MRI scans which were previously delineated by expert radiologists and neurosurgeons. We then tested the model against lumbar MRI scans for 140 patients who had undergone lumbar spine MRI at our institution (60 of whom ultimately underwent surgery, and 80 of whom did not). The model generated automated segmentations of the lumbar spinal canals and calculated a maximum degree of spinal stenosis for each patient, which served as our biomarker for surgical pathology warranting expert consultation. RESULTS: The machine learning model correctly predicted surgical candidacy (ie, whether patients ultimately underwent lumbar spinal decompression) with high accuracy (area under the curve = 0.88), using only imaging data from lumbar MRI scans. CONCLUSION: Automated interpretation of lumbar MRI scans was sufficient to correctly determine surgical candidacy in nearly 90% of cases. Given that a significant proportion of referrals placed for spine surgery evaluation fail to meet criteria for surgical intervention, our model could serve as a valuable tool for patient triage and thereby address some of the inefficiencies within the outpatient surgical referral process.

Eigenrank by committee: Von-Neumann entropy based data subset selection and failure prediction for deep learning based medical image segmentation.

Manual delineation of anatomy on existing images is the basis of developing deep learning algorithms for medical image segmentation. However, manual segmentation is tedious. It is also expensive because clinician effort is necessary to ensure correctness of delineation. Consequently most algorithm development is based on a tiny fraction of the vast amount of imaging data collected at a medical center. Thus, selection of a subset of images from hospital databases for manual delineation - so that algorithms trained on such data are accurate and tolerant to variation, becomes an important challenge. We address this challenge using a novel algorithm. The proposed algorithm named 'Eigenrank by Committee' (EBC) first computes the degree of disagreement between segmentations generated by each DL model in a committee. Then, it iteratively adds to the committee, a DL model trained on cases where the disagreement is maximal. The disagreement between segmentations is quantified by the maximum eigenvalue of a Dice coefficient disagreement matrix a measure closely related to the Von Neumann entropy. We use EBC for selecting data subsets for manual labeling from a larger database of spinal canal segmentations as well as intervertebral disk segmentations. U-Nets trained on these subsets are used to generate segmentations on the remaining data. Similar sized data subsets are also randomly sampled from the respective databases, and U-Nets are trained on these random subsets as well. We found that U-Nets trained using data subsets selected by EBC, generate segmentations with higher average Dice coefficients on the rest of the database than U-Nets trained using random sampling (p < 0.05 using t-tests comparing averages). Furthermore, U-Nets trained using data subsets selected by EBC generate segmentations with a distribution of Dice coefficients that demonstrate significantly (p < 0.05 using Bartlett's test) lower variance in comparison to U-Nets trained using random sampling for all datasets. We believe that this lower variance indicates that U-Nets trained with EBC are more robust than U-Nets trained with random sampling.

Comparison of Clinical Outcomes Stratified by Target Delineation for Patients Undergoing Stereotactic Body Radiotherapy for Spinal Metastases.

OBJECTIVE: Stereotactic body radiotherapy (SBRT) is an effective treatment of spinal metastases in the vertebral body. However, variation has existed between practitioners regarding the appropriate target delineation. As such, we compared the tumor control, rates of compression fractures, and pain control for patients who had undergone SBRT for spinal metastases to either the lesion only (LO) or the full vertebral body (FVB). METHODS: A total of 126 spinal metastases in 84 patients had received single-fraction SBRT from January 2009 to February 2015. Of the 126 lesions, 36 (29%) were in the FVB group and 90 were in the LO group. The SBRT plans were reviewed to determine the treatment volume. Odds ratios were used to compare the rates of compression fracture and local failure. Regression analysis was performed to identify the predictors of outcome. RESULTS: A total of 5 failures had occurred in the FVB group and 14 in the LO group; however, the difference was not statistically significant (P = 0.5). No difference was found in pain reduction between the 2 groups (P = 0.9). Seven post-treatment compression fractures occurred in the LO group and four in the FVB group; however, the difference was not statistically significant (P = 0.6). The minimum dose to the planning target volume, patient age, and planning target volume size were the only significant factors predicting for local failure, vertebral body fracture, and pain control, respectively. CONCLUSIONS: Given that we found no difference in tumor control, pain reduction, or fracture rate between patients treated to the FVB versus the. LO, it might be reasonable to consider SBRT to the LO for select patients.

United States Medical Licensing Examination step 2 scores do not predict American Board of Neurological Surgery scores: A single-institution experience.

OBJECTIVES: The neurosurgery residency match is becoming increasingly competitive, with numerous factors being considered as part of the application. We aim to determine whether USMLE Step 2 scores were a significant predictor of neurosurgery board performance. PATIENTS AND METHODS: Residents who entered a neurological surgery residency program at a single academic institution during 2000-2017 provided scores for all ABNS attempts, USMLE Step 1 and Step 2 scores. Data were deidentified and analyzed for correlation and regression. Pearson's correlation coefficients were determined. RESULTS: USMLE Step 1, Step 2, and maximum ABNS scores were all normally distributed. Step 1 and Step 2 scores were less variable than ABNS scores. USMLE Step 2 and residents' best ABNS written examination scores were not correlated (Pearson Correlation of 0.228 with a 2-tailed significance of 0.272). No outliers were present. When comparing USMLE Step 2 scores with year in residency at which residents scored over 300 on the ABNS written examination, Pearson correlation was -0.500 (p = .015). A simple linear regression was calculated using Step 2 scores to predict the passing year of ABNS written examination (F(1,14) = 6.984, p = .015, R2 = 0.25). CONCLUSION: Although other studies have found correlations between USMLE Step 2 scores and performance before graduating medical school and during residency for other specialties, this is the first study comparing USMLE Step 2 scores with the ABNS written examination scores of neurosurgical residents. Our data showed that USMLE Step 2 was not a reliable predictor of ABNS written examination scores.

Clinical and surgical management of holocervical spinal cord ependymomas.

BACKGROUND: Spinal ependymomas are rare tumors of the central nervous system, and those spanning the entire cervical spine are atypical. Here, we present two unusual cases of holocervical (C1-C7) spinal ependymomas. CASE DESCRIPTION: Two patients, a 32-year-old female and a 24-year-old male presented with neck pain, motor, and sensory deficits. Sagittal MRI confirmed hypointense lesions on T1 and hyperintense regions on T2 spanning the entire cervical spine. These were accompanied by cystic cavities extending caudally into the thoracic spine and rostrally to the cervicomedullary junction. Both patients underwent gross total resection of these lesions and sustained excellent recoveries. CONCLUSION: Two holocervical cord intramedullary ependymomas were safely and effectively surgically resected without incurring significant perioperative morbidity.

Natural history of the aging spine: a cross-sectional analysis of spinopelvic parameters in the asymptomatic population.

OBJECTIVE: There have been numerous studies demonstrating increased pain and disability when patients' spinopelvic parameters fall outside of certain accepted ranges. However, these values were established based on patients suffering from spinal deformities. It remains unknown how these parameters change over a lifetime in asymptomatic individuals. The goal of this study was to define a range of spinopelvic parameters from asymptomatic individuals. METHODS: Sagittal scoliosis radiographs of 210 asymptomatic patients were evaluated. All measurements were reviewed by 2 trained observers, supervised by a trained clinician. The following parameters and relationships were measured or calculated: cervical lordosis (CL), thoracic kyphosis (TK), lumbar lordosis (LL), pelvic incidence (PI), sagittal vertical axis (SVA), cervical SVA (cSVA), and T1 slope, TK/LL, truncal inclination, pelvic tilt (PT), LL-PI, LL/PI, and T1 slope/PI. Patients were stratified by decade of life, and regression analysis was performed to delineate the relationship between each consecutive age group and the aforementioned parameters. RESULTS: Cervical lordosis (R2 = 0.61), thoracic kyphosis (R2 = 0.84), SVA (R2 = 0.88), cSVA (R2 = 0.51), and T1 slope (R2 = 0.77) all increase with age. Truncal inclination (R2 = 0.36) and T1 slope/CL remain stable over all decades (R2 = 0.01). LL starts greater than PI, but in the 6th decade of life, LL becomes equal to PI and in the 7th decade becomes smaller than PI (R2 = 0.96). The ratio of TK/LL is stable until the 7th decade of life (R2 = 0.81), whereas PT is stable until the 6th decade (R2 = 0.92). CONCLUSIONS: This study further refines the generally accepted LL = PI + 10° by showing that patients under the age of 50 years should have more LL compared to PI, whereas after the 5th decade the relationship is reversed. SVA was not as sensitive across age groups, exhibiting a marked increase only in the 7th decade of life. Given the reliable increase of CL with age, and the stability of T1 slope/CL, this represents another important relationship that should be maintained when performing cervical deformity/fusion surgery. This study has important implications for evaluating adult patients with spinal deformities and for establishing corrective surgical goals.

Cangrelor Bridging Within 6 Months of a Drug-Eluting Stent in a Patient for Emergency Cervical Spine Surgery: A Case Report.

The recommended duration of dual antiplatelet therapy after drug-eluting stent placement presents a dilemma for patients with recent stenting who require urgent or emergency noncardiac surgery. We present the case of a patient with recent drug-eluting stent placement (<6 months) on dual antiplatelet therapy who underwent successful emergency cervical spine surgery with antiplatelet therapy bridged using cangrelor, an intravenous P2Y12 inhibitor antiplatelet agent. Our experience illustrates the multidisciplinary approach to a patient with high thrombotic and bleeding risk who underwent neurosurgery off both aspirin and a P2Y12 inhibitor.

Quantitative Analysis of Neural Foramina in the Lumbar Spine: An Imaging Informatics and Machine Learning Study.

PURPOSE: To use machine learning tools and leverage big data informatics to statistically model the variation in the area of lumbar neural foramina in a large asymptomatic population. MATERIALS AND METHODS: By using an electronic health record and imaging archive, lumbar MRI studies in 645 male (mean age, 50.07 years) and 511 female (mean age, 48.23 years) patients between 20 and 80 years old were identified. Machine learning algorithms were used to delineate lumbar neural foramina autonomously and measure their areas. The relationship between neural foraminal area and patient age, sex, and height was studied by using multivariable linear regression. RESULTS: Neural foraminal areas correlated directly with patient height and inversely with patient age. The associations involved were statistically significant (P < .01). CONCLUSION: By using machine learning and big data techniques, a linear model encoding variation in lumbar neural foraminal areas in asymptomatic individuals has been established. This model can be used to make quantitative assessments of neural foraminal areas in patients by comparing them to the age-, sex-, and height-adjusted population averages.© RSNA, 2019Supplemental material is available for this article.

Cell Counting and Segmentation of Immunohistochemical Images in the Spinal Cord: Comparing Deep Learning and Traditional Approaches.

Estimation of cell nuclei in images stained for the c-fos protein using immunohistochemistry (IHC) is infeasible in large image sets. Use of multiple human raters to increase throughput often creates variance in the data analysis. Machine learning techniques for biomedical image analysis have been explored for cell-counting in pathology, but their performance on IHC staining, especially to label activated cells in the spinal cord is unknown. In this study, we evaluate different machine learning techniques to segment and count spinal cord neurons that have been active during stepping. We present a qualitative as well as quantitative comparison of algorithmic performance versus two human raters. Quantitative ratings are presented with cell-count statistics and Dice (DSI) scores. We also show the degree of variability between multiple human raters' segmentations and observe that there is a higher degree of variability in segmentations produced by classic machine learning techniques (SVM and Random forest) as compared to the newer deep learning techniques. The work presented here, represents the first steps towards addressing the analysis time bottleneck of large image data sets generated by c-fos IHC staining techniques, a task that would be impossible to do manually.

ORCHESTRAL FULLY CONVOLUTIONAL NETWORKS FOR SMALL LESION SEGMENTATION IN BRAIN MRI.

White matter (WM) lesion identification and segmentation has proved of clinical importance for diagnosis, treatment and neurological outcomes. Convolutional neural networks (CNN) have demonstrated their success for large lesion load segmentation, but are not sensitive to small deep WM and sub-cortical lesion segmentation. We propose to use multi-scale and supervised fully convolutional networks (FCN) to segment small WM lesions in 22 anemic patients. The multiple scales enable us to identify the small lesions while reducing many false alarms, and the multi-supervised scheme allows a better management of the unbalanced data. Compared to a single FCN (Dice score ~0.31), the performance on the testing dataset of our proposed networks achieved a Dice score of 0.78.

Autonomous Trajectory Planning for External Ventricular Drain Placement.

BACKGROUND: External ventricular drain (EVD) placement is the most frequently performed neurosurgical procedure for management of various conditions including hydrocephalus, traumatic brain injury, and stroke. State-of-the-art computational pattern recognition techniques could improve the safety and accuracy of EVD placement. Placement of the Kocher's point EVD is the most common neurosurgical procedure which is often performed in urgent conditions. OBJECTIVE: To present the development of a novel computer algorithm identifying appropriate anatomy and autonomously plan EVD placement on clinical computed tomography (CT) scans. METHODS: The algorithm was tested on 2 data sets containing 5-mm slice noncontrast CT scans. The first contained images of 300 patients without significant intracranial pathology (normal), the second of 43 patients with significant acute intracranial hemorrhage. Automated planning was performed by custom 2-tiered heuristic with run-time template selection in combination with refinement using nonlinear image registration. RESULTS: Automated EVD planning was accurate in 297 of 300 normal and 41 of 43 patient cases. In the normal data set, mean distance between Kocher's point and the ipsilateral foramen of Monro was 63 ± 3.1 mm in women and 65 ± 6.5 mm in men (P = .0008). Trajectory angle with respect to the sagittal plane was 91 ± 6° in women and 90 ± 6° in men (obtuse posterior) (P = .15); to the coronal plane, 85 ± 6° and 86 ± 5° in women and men (P = .12), respectively (acute lateral). CONCLUSION: A combination of linear and nonlinear image registration techniques accurately planned EVD trajectory in 99% of normal scans and 95% of scans with significant intracranial hemorrhage.

Multi-Parameter Ensemble Learning for Automated Vertebral Body Segmentation in Heterogeneously Acquired Clinical MR Images.

The development of quantitative imaging biomarkers in medicine requires automatic delineation of relevant anatomical structures using available imaging data. However, this task is complicated in clinical medicine due to the variation in scanning parameters and protocols, even within a single medical center. Existing literature on automatic image segmentation using MR data is based on the analysis of highly homogenous images obtained using a fixed set of pulse sequence parameters (TR/TE). Unfortunately, algorithms that operate on fixed scanning parameters do not avail themselves to real-world daily clinical use due to the existing variation in scanning parameters and protocols. Thus, it is necessary to develop algorithmic techniques that can address the challenge of MR image segmentation using real clinical data. Toward this goal, we developed a multi-parametric ensemble learning technique to automatically detect and segment lumbar vertebral bodies using MR images of the spine. We use spine imaging data to illustrate our techniques since low back pain is an extremely common condition and a typical spine clinic evaluates patients that have been referred with a wide range of scanning parameters. This method was designed with special emphasis on robustness so that it can perform well despite the inherent variation in scanning protocols. Specifically, we show how a single multi-parameter ensemble model trained with manually labeled T2 scans can autonomously segment vertebral bodies on scans with echo times varying between 24 and 147 ms and relaxation times varying between 1500 and 7810 ms. Furthermore, even though the model was trained using T2-MR imaging data, it can accurately segment vertebral bodies on T1-MR and CT, further demonstrating the robustness and versatility of our methodology. We believe that robust segmentation techniques, such as the one presented here, are necessary for translating computer assisted diagnosis into everyday clinical practice.